modbender LLM Council (with Installer)
LLM Council — ask one question to many models, let them critique each other, get a synthesized chairman answer.
This skill is the fastest way to run it: one command installs dependencies, configures credentials, and launches both backend and frontend. No manual setup, no API key prompts.
OpenClaw-native: Credentials resolve automatically from OpenClaw config or workspace .env. Falls back to the local OpenClaw gateway (port 18789) if no OpenRouter key is found.
Two Ways to Use LLM Council
| Mode | Best For | Command |
|---|---|---|
| Quick answer | Fast decisions, mobile, casual questions | /council "Your question" (requires ask-council skill) |
| Full discussion | Deep research, exploring disagreements, seeing all model responses | /install-llm-council then open browser at :5173 |
Slash Command
/install-llm-council [--mode auto|dev|preview] [--dir PATH]
When the user says /install-llm-council, run:
bash ~/.openclaw/skills/install-llm-council/install.sh
The script will:
- Resolve credentials — env var → workspace
.env→ OpenClaw local gateway (no prompt ever) - Clone or pull
https://github.com/jeadland/llm-councilto~/workspace/llm-council uv sync— Python backend dependenciesnpm ci— frontend dependencies- Write
.env— API key/URL for OpenRouter direct or OpenClaw gateway mode - Start app — uses hardened
start.shwith mode-aware startup and health checks - Auto-handle port conflicts — selects safe fallback ports when defaults are busy
- Print practical access URLs — Caddy route and common direct fallbacks
Flags
| Flag | Default | Description |
|---|---|---|
--mode auto |
auto |
Detect Caddy on :5173 and prefer preview mode; otherwise dev mode |
--mode dev |
— | Run Vite dev server (hot reload, port 5173 default) |
--mode preview |
— | Build + run Vite preview (port 4173 default) |
--dir PATH |
~/workspace/llm-council |
Override clone directory |
Credential Resolution (OpenClaw-native)
The installer never prompts for API keys. It resolves credentials in this order:
- Environment —
OPENROUTER_API_KEYalready exported - Workspace
.env—~/.openclaw/workspace/.envcontainsOPENROUTER_API_KEY=... - OpenClaw gateway — reads
~/.openclaw/openclaw.json→gateway.auth.token+gateway.port- Sets
OPENROUTER_API_URL=http://127.0.0.1:<port>/v1/chat/completionsin.env - Uses the gateway token as the bearer key (OpenAI-compatible endpoint)
- Sets
Ports
| Service | Port | Notes |
|---|---|---|
| Backend (FastAPI) | 8001 | Always |
| Frontend dev | 5173 | --mode dev (default) |
| Frontend preview | 4173 | --mode preview |
Files
| File | Purpose |
|---|---|
SKILL.md |
This file — skill documentation |
install.sh |
Main one-shot installer/launcher |
stop.sh |
Stop background services |
status.sh |
Check if services are running |
pids |
Saved PIDs for background processes |
Agent Instructions
When user says /install-llm-council or "install llm-council" or "start llm council":
bash ~/.openclaw/skills/install-llm-council/install.sh
Report back the access URL from the script output (e.g. http://10.0.1.X:5173).
To stop:
bash ~/.openclaw/skills/install-llm-council/stop.sh
To check status:
bash ~/.openclaw/skills/install-llm-council/status.sh
Example Output
✅ LLM Council installed and running!
Mode: dev
API: openrouter
Backend: http://127.0.0.1:8001
Frontend: http://10.0.1.42:5173
Stop: bash ~/.openclaw/skills/install-llm-council/stop.sh
Status: bash ~/.openclaw/skills/install-llm-council/status.sh